git.ipfire.org Git - thirdparty/kernel/stable-queue.git/blob

   1 From 96b6e3e31b8600d1c9544f8e0cfdfe205c2db54f Mon Sep 17 00:00:00 2001
   2 From: Sasha Levin <sashal@kernel.org>
   3 Date: Mon, 2 Nov 2020 15:31:27 -0500
   4 Subject: ring-buffer: Fix recursion protection transitions between interrupt
   5  context
   6
   7 From: Steven Rostedt (VMware) <rostedt@goodmis.org>
   8
   9 [ Upstream commit b02414c8f045ab3b9afc816c3735bc98c5c3d262 ]
  10
  11 The recursion protection of the ring buffer depends on preempt_count() to be
  12 correct. But it is possible that the ring buffer gets called after an
  13 interrupt comes in but before it updates the preempt_count(). This will
  14 trigger a false positive in the recursion code.
  15
  16 Use the same trick from the ftrace function callback recursion code which
  17 uses a "transition" bit that gets set, to allow for a single recursion for
  18 to handle transitions between contexts.
  19
  20 Cc: stable@vger.kernel.org
  21 Fixes: 567cd4da54ff4 ("ring-buffer: User context bit recursion checking")
  22 Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
  23 Signed-off-by: Sasha Levin <sashal@kernel.org>
  24 ---
  25  kernel/trace/ring_buffer.c | 54 +++++++++++++++++++++++++++++++-------
  26  1 file changed, 44 insertions(+), 10 deletions(-)
  27
  28 diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
  29 index b9b71e7fb6979..8082328eb01a4 100644
  30 --- a/kernel/trace/ring_buffer.c
  31 +++ b/kernel/trace/ring_buffer.c
  32 @@ -416,14 +416,16 @@ struct rb_event_info {
  33
  34  /*
  35   * Used for which event context the event is in.
  36 - *  NMI     = 0
  37 - *  IRQ     = 1
  38 - *  SOFTIRQ = 2
  39 - *  NORMAL  = 3
  40 + *  TRANSITION = 0
  41 + *  NMI     = 1
  42 + *  IRQ     = 2
  43 + *  SOFTIRQ = 3
  44 + *  NORMAL  = 4
  45   *
  46   * See trace_recursive_lock() comment below for more details.
  47   */
  48  enum {
  49 +       RB_CTX_TRANSITION,
  50         RB_CTX_NMI,
  51         RB_CTX_IRQ,
  52         RB_CTX_SOFTIRQ,
  53 @@ -2553,10 +2555,10 @@ rb_wakeups(struct ring_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer)
  54   * a bit of overhead in something as critical as function tracing,
  55   * we use a bitmask trick.
  56   *
  57 - *  bit 0 =  NMI context
  58 - *  bit 1 =  IRQ context
  59 - *  bit 2 =  SoftIRQ context
  60 - *  bit 3 =  normal context.
  61 + *  bit 1 =  NMI context
  62 + *  bit 2 =  IRQ context
  63 + *  bit 3 =  SoftIRQ context
  64 + *  bit 4 =  normal context.
  65   *
  66   * This works because this is the order of contexts that can
  67   * preempt other contexts. A SoftIRQ never preempts an IRQ
  68 @@ -2579,6 +2581,30 @@ rb_wakeups(struct ring_buffer *buffer, struct ring_buffer_per_cpu *cpu_buffer)
  69   * The least significant bit can be cleared this way, and it
  70   * just so happens that it is the same bit corresponding to
  71   * the current context.
  72 + *
  73 + * Now the TRANSITION bit breaks the above slightly. The TRANSITION bit
  74 + * is set when a recursion is detected at the current context, and if
  75 + * the TRANSITION bit is already set, it will fail the recursion.
  76 + * This is needed because there's a lag between the changing of
  77 + * interrupt context and updating the preempt count. In this case,
  78 + * a false positive will be found. To handle this, one extra recursion
  79 + * is allowed, and this is done by the TRANSITION bit. If the TRANSITION
  80 + * bit is already set, then it is considered a recursion and the function
  81 + * ends. Otherwise, the TRANSITION bit is set, and that bit is returned.
  82 + *
  83 + * On the trace_recursive_unlock(), the TRANSITION bit will be the first
  84 + * to be cleared. Even if it wasn't the context that set it. That is,
  85 + * if an interrupt comes in while NORMAL bit is set and the ring buffer
  86 + * is called before preempt_count() is updated, since the check will
  87 + * be on the NORMAL bit, the TRANSITION bit will then be set. If an
  88 + * NMI then comes in, it will set the NMI bit, but when the NMI code
  89 + * does the trace_recursive_unlock() it will clear the TRANSTION bit
  90 + * and leave the NMI bit set. But this is fine, because the interrupt
  91 + * code that set the TRANSITION bit will then clear the NMI bit when it
  92 + * calls trace_recursive_unlock(). If another NMI comes in, it will
  93 + * set the TRANSITION bit and continue.
  94 + *
  95 + * Note: The TRANSITION bit only handles a single transition between context.
  96   */
  97
  98  static __always_inline int
  99 @@ -2597,8 +2623,16 @@ trace_recursive_lock(struct ring_buffer_per_cpu *cpu_buffer)
 100         } else
 101                 bit = RB_CTX_NORMAL;
 102
 103 -       if (unlikely(val & (1 << bit)))
 104 -               return 1;
 105 +       if (unlikely(val & (1 << bit))) {
 106 +               /*
 107 +                * It is possible that this was called by transitioning
 108 +                * between interrupt context, and preempt_count() has not
 109 +                * been updated yet. In this case, use the TRANSITION bit.
 110 +                */
 111 +               bit = RB_CTX_TRANSITION;
 112 +               if (val & (1 << bit))
 113 +                       return 1;
 114 +       }
 115
 116         val |= (1 << bit);
 117         cpu_buffer->current_context = val;
 118 --
 119 2.27.0
 120